Research in a sunny Saturday

Found many ways have been tried, the only and most different part is that non of the work has tried query-based ordering. So give question an initial try.

Reading a real set of docs and identify the pattern

+

Final Data Location for traning:

DUC05 /cs/natlang-d/duc2006/data/Archive/DUC05_SQuASH_v36FinalTrain
DUC06 /cs/natlang-d/duc2006/data/Archive/DUC06_SQuASH_v36Submission